Ingre, M., Sorjonen, K., & Nilsonne, G. (2019, February 24). A cocktail fallacy in research with composite variables: a critical review of job strain, effort–reward imbalance (ERI) and body mass index (BMI). https://doi.org/10.31234/osf.io/9hyxj
To estimate a combined effect of two different exposures, the standard method is to include both exposures as independent variables in a statistical model, together with an interaction term. Journals such as Epidemiology recommend this method as best practice in their instructions for authors. However, in occupational stress research it is common to combine two exposures into a single composite variable, and then use associations observed on that variable to claim support for theories implying an interaction (or a combined additive effect). Here we provide a non-technical illustration of how such composite variable models lead to a logical fallacy, when they are used to try to argue support for theories such as job strain and effort–reward imbalance (ERI). We discuss different types of composite variables, illustrate their susceptibility to bias, and show why researchers should be critical when interpreting associations observed on the body mass index (BMI). We also present the proper statistical models that should be used when estimating associations in job strain and effort–reward imbalance or similar research.